The AcrB Efflux Pump : Conformational Cycling and Peristalsis Lead to Multidrug Resistance

(1)

729

The AcrB Efflux Pump: Conformational Cycling and Peristalsis Lead to Multidrug Resistance

Markus A. Seegerl

,2,3, Kay Diederichs4, Thomas Eicherl

, Lorenz Brandstatterl

, Andre Schiefner4 ,5, Franc;:ois Verreyl and Klaas M. POSI,6,*

'Institute of Physiology and Zurich Centre for Integrative Human Physiology (ZIHP), University of Zurich, Winterthur- erstrasse 190, CH-8057 Ziirich, Switzerland; 21nstitute of Microbiology, Swiss Federal Institute of Technology (ET/-/) , Wolfgang-Pauli-Strasse iO, C/-1-8093 Ziirich, Switzerland; JCw'rent address: Department of Pharmacology, University of Cambridge, Tennis Courl Road, Cambridge CB2 I PD, United Kingdom; 4Department of Biology, University of Kon- stanz, Universitiitsstrasse 10, M647, D-78457 Konstanz, Germany; 5Present address: Department of Molecular Biology, BCC206, The Scripps Research Institule (TSRl), 10550 North Torrey Pines Road, La Jolla, CA 92037, USA and Current Address: institute of BiochemistlY and Cluster of Excellence Macromolecular Complexes, Johann Wolfang Goethe- Universily, Max-von-Laue-Str. 9, D-60438, Frankfurt am Main, Germany

Abstract: Antimicrobial resistance of human pathogenic bacteria is an emerging problem for global public health. This resistance is often associated with the overproduction of membrane transport proteins that are capable to pump chemotherapeutics, antibiotics, detergents, dyes and organic solvents out of the cell. In Gram-negative bacteria such as Es- cherichia coli and Pseudomonas aeruginosa, tripartite multidrug efflux systems extrude a large variety of cytotoxic substances from the cell membrane directly into the medium bypassing the periplasm and the outer membrane. In E. coli, the tripartite efflux system AcrA/AcrBrrolC is the pump in charge of the efflux of multiple antibiotics, dyes, bile salts and detergents. The trimeric outer membrane factor (OMF) TolC fonns a ~-barrel pore in the outer membrane and exhibits a long periplasmic a-helical conduit. The periplasmic membrane fusion protein (MFP) AcrA serves as a linker between TolC and the trimeric resistance nodulation cell division ^(RND)pump AcrB, located in the inner membrane acting as a proton/drug antiporter.

The newly elucidated asymmetric structure of trimeric AcrB reveals three different monomer conformations representing consecutive states in a transport cycle. The monomers show tunnels with occlusions at different sites leading from the lat- etal side through the periplasmic porter (pore) domains towards the I1l11nel of the trimer and ToIC. The structural changes create a hydrophobic pocket in one monomer, which is not present in the other two monomers. Minocyclin and doxorubicin, both AcrB substrates, specifically bind to this pocket substantiating its role as drug binding pocket. The energy transduction from the proton motive force into drug efflux includes proton binding in (and release from) the transmembrane part. The confonnational changes observed within a triad of essential, titratable residues (Asp407/Asp40S/Lys940) resid- ing in the hydrophobic transmembrane domain appear to be transduced by transmembrane helix S and associated with the confonnational changes seen in the periplasmic domain.

From the asymmetric structure a possible peristaltic pump transport mechanism based on a functional rotation of the AcrB trimer has been postulated. The novel transport model merges .Iardetzky's alternate access pump mechanism with the ro- tating site catalysis of F I F 0 ATPase and suggests a working hypothesis for the transport mechanism 0 f RND transporters in general.

INTRODUCTION Drug Efflux Transl)Orters

Based on a systematic genome-wide analysis of membrane transport systems, so lute transporters were identi fied and classified into about 550 families and sLiperfamilies.

Five superfam il ies comprising primary and secondary active drug transporters have been described and classified according to the approved transporter classification system [I).

These are the ATP binding cassette (ABC) superfamily (TC#3.A.I), the major facilitator superfamily (MFS) (TC#2.A.I), the resistance/nodulation/division (RND) superfamily (TC#2.A.6), the drug/metabolite transporter (DMT)

'Address correspondence to this author at the Institute of Physiology, Uni- versity of Zurich, Winterlhurerstrasse 190, C1-1-8057 ZOrich, Switzerland;

Tel: +41-44-6355046; Fax: +41-44-635 6814;

E-mail: kmpos@access.uzh.ch

superfamily (TC#2.A.7) and the multi antimicrobial extrusion (MATE) family (TC#2.A.66) [2]. ABC drug transporters are primary active transporters and in eukaryotes they playa major role in extrusion of noxious substances out of the cell. In addition, transport of chemotherapeutics out of tumour cells is mediated by ABC-transporters (e.g. Pgp) and as a consequence causes failure of cancer therapy. LmrA of the Gram-positive bacterium Lactococcus lactis is a well- studied prokaryotic ABC drug transporter, which transports different positively charged drugs [3). Recently, a structure of its close homologue, Sav 1866 of Staphylococcus aureus, was solved by X-ray crystallography at 3

A

resolution (4).

The drug-trans locating members of the other four superfamilies all belong to the class of electrochemical potential- driven transporters (Fig. 1). Among the members of the MFS multidrug resistance transporters, MdfA is functionally well characterised and extrudes positively charged and neutral First publ. in: Current Drug Targets ; 9 (2008), 9. - S. 729-749

Konstanzer Online-Publikations-System (KOPS) URN: http://nbn-resolving.de/urn:nbn:de:bsz:352-opus-119426

(2)

Periplasm

EmrO(MFS) EmrE (OMT) AcrB (RNO)

Fig. (I). Secondary active multidrug transporters. Structme models from members of the major facilitator superfamily (MFS), the drug/metabolite transporter (OMT) superfamily and the resistance nodulation division (RNO) superfamily. From left to right: MFS: EmrO (POB entry: 2GFP) OMT: EmrE (POB entry: 3B62) and AcrB (POB entries: IIWG, 2GIF). Representative structures from 3 out of 4 secondary active multidrug transporter superfamilies are shown. There is no clystal structure available from the Fourth Family, the multi antimicrobial extrusion (MATE) family (e.g. NorM).

drugs [5]. The structure of its homologue EmrD has been recently solved at 3.5

A

resolution [6). Its structure appears similar to the structures of other MFS superfamily members such as LacY and GlpT (7, S]. EmrE has been studied on the functional and structural level and belongs to the small multidrug resistance family (SMR) which is part of the drug/metabolite transporter (DMT) superfamily [9, 10]. One member of the multi antimicrobial extrusion (MATE) family, NorM, is described as a sodium-ion driven efflux pump of Vibrio parahaemolyticus [II]. Drug transporting members of the RND superfamily, including AcrB, are among the most intensively studied transport systems.

The RND Superfamily

Members of the RND superfamily (TC#2.A.6) are found ubiquitously in bacteria, archaea and eukaryotes. They mainly function as proton/substrate anti porters and belong to eight phylogenetic families [2]. Clustering patterns in these families of the RND superfamily correlate with substrate specificity. For example, the heavy metal efflux (HME) family (TC#2.A.6.1) catalyzes export of heavy metal ions (Zn²+,

CU²) and the hydrophobe/amphiphile efflux-I (HAE I) family (TC#2.A.6.2) catalyzes the export of multiple drugs. The best characterized example of the HME family is CzcA from Ralstonia eutropha, acting as a heavy metal-ion/proton antiporter [12]. AcrB of Escherichia coli and MexB of Pseu- domonas aeruginosa are the best studied members of the I-lAEI family [13, 14]. Human NPCI being involved in lipid transport through the membrane and associated with the Niemann-Pick C I disease belongs to the eukaryotic sterol transporter (EST) family (TC#2.A.6.6) [IS]. Most RND superfam ily transporters consist of a single large polypeptide chain (700-1300 amino acid residues) and are predicted to span the membrane 12 times with two large periplasm ic do-

mains located between transmembrane helices I and 2 (TM I and TM2) and between TM7 and TMS.

RND/MFP/OMF Efflux Systems ill Escherichia coli The RND members of the HME and HAEI families function in conjunction with a membrane fusion protein (MFP; TC #S.A.I) and an outer membrane factor (OMF; TC

# I.B.17) [2] (Fig. 2). The E. coli genome encodes 4 MFPs and 6 RND pumps. The genes acrAB, acrEF and mdtEF coding for the MFP and RND members, respectively, are co- transcribed. On the other hand, the gene encoding the RND pump AcrD is found isolated on the chromosome. The mdtABCD operon consists of mdtA encoding a MFP, mdtB and mdtC encoding two RND pumps and mdtD encoding a MFS transporter. MFPs and RND proteins which are coded by co-transcribed genes form functional entities. However, some of the MFPs can form functional complexes with RND pumps in spite of the fact that the encoding genes are transcribed independently. Table I summarizes all known functional partnerships between the MFPs and RND pumps in E.

coli. Remarkably, all pairs of MFPs and RND proteins in E.

coli share TolC as outer membrane factor which is encoded independently elsewhere on the chromosome [16, 17]. Sys- tematic deletion of E. coli genes coding for RND, MFS, SMR and ABC transporter family members as well as MFPs and OMFs revealed that only the tripartite AcrA/AcrB/TolC efflux system contributes to the observed antibiotic resistance in E. coli wild type strains, whereas other transporters (including RND members) appear to playa minor role under standard laboratory conditions [IS]. However, if the genes

acrD, acrEF, mdtEF and mdfABCD encoding constituents of

the AcrA/AcrD/ToIC, AcrE/AcrF/ToIC, MdtE/MdtF/TolC and MdtA/MdtB/MdtC/TolC efflux systems, respectively, were overproduced in an acrB deficient background, partial

(3)

outside

periplasm

cytoplasm

T

outer membrane channel

AcrA membrane fusion protein linker between TolC and AcrB

AcrB multiple resistance antiporter driven by proton motive force

energy module and substrate specificity determinant

Fig. (2). Schematic drawing based on the X-ray structures of the tripartite multidrug efflux system AcrAB-ToIC of Gram-negative E. coli.

AcrB (RND component, in blue color) resides in the inner membrane and is responsible for substrate recognition/selection and energy transduction. Drugs are captured at the inner or outer leaflet of the inner membrane and extruded in a coupled exchange with protons. TolC (OMF component, yellow) forms a pore in the outer membrane which is extended by a long periplasmic conduit. AcrA (MFP component, red) me- diates contact between AcrB and ToIC. The presence of all three components is essential for the MDR phenotype.

recovery of the resistance for some of the tested drugs was observed [19]. It appears that the AcrAIAcrB/TolC system is constitutively expressed at a low level, whereas the other RND/MFP efflux systems are tightly repressed in E. coli.

The acr locus coding for acrA and acrB was mapped as early as in 1965 [20], when acriflavine sensitive E. coli strains were genetically analyzed. In addition, the mutants were shown to be susceptible to other dyes such as crystal violet and pyronine B. However, il took 28 years until Nikaido and co-workers postulated in 1993 that AcrA and AcrB function as components of an efflux pump system [21].

RND/MFP/OMF Efflux Systems in Pseudomonas aerugi- nosa

At around the same time as the research from the laboratory of H. Nikaido was published, Poole and co-workers [26]

postulated that proteins encoded by genes of the P. aerugi- nasa operon mexAB-oprM constitute a multidrug eftlux

pump, where MexA is denoted the MFP, MexB the RND pump and OprM the OMF. Other than with TolC of E. coli, OprM is encoded by a gene located on the same operon as the other constituents of the P. aeruginosa tripartite efflux pump. Currently there are numerous tripartite RND/MFPI OMF efflux systems characterized in P. aeruginosa named MexC/MexD/OprJ, MexE/MexFIOprN, MexX/MexY/OprM, MexJ/MexK/OprM, MexH/MexIlOpmD and MexV/MexWI OprM. The coding genes of some of these tripartite systems are located on the same operon, whereas in others the corresponding OMF gene is located at a transcriptionally inde- pendent position on the chromosome (reviewed in [27]).

Substrates of the Tripartite RND/MFP/OMF Efflux Sys- tems

RND/MFP/OMF type efflux systems have been shown to exhibit a wide substrate specificity and extrude a broad range of chemically unrelated compounds including dyes, deter-

Table I. Interacting Combinations of RND and MFP Components of E. coli. All RND/MFP Pairs Exclusively Recruit TolC as OMF. n) According to 1221, b) According to 1231, c) According to 1241, d) According to 1251

RND Component

AcrB AerD AcrF(EnvD) MdtF(YhiV) MdtB(YegN) MdtC(YegO) MdtBC

AcrA yes') yes") yes") yes') no') no') n.d.

AcrE (EnvC) yes') n.d. yes") n.d. n.d. n.d. n.d.

MFI' COIllPonent

MdtE (YhiU) ^no^") n.d. n.d. yes') n.d. n.d. n.d.

MdtA(YegM) no') n.d. n.d. n.d. ^no^h),c) no b), yes^C) yesh).c)

(4)

gents, bile salts, different classes of antibiotics and even organic solvents (Fig. 3) [\9,28-30]. The substrate molecules are structurally very diverse, can be charged or neutral and drastically differ in molecular weight. Some substrates are clearly amphiphilic i.e. having a large hydrophobic part and a small polar, often charged hydrophilic part (e.g. taurocholate, dodecylsulfate, oxacillin), whereas other substrates con- tain a delocalized charge within an aromatic environment (e.g ethidium, berberine, acriflavine). Yet others are simple

""

^.

CQOCH2CH3

( ""

6"

~

><:::::~H 0

H2N #

7 #

^...# ^NH2

\ CH3 CH,

N

aliphatic or alicyclic compounds (e.g. hexane, cyclohexane).

The AcrAl AcrB/TolC efflux system transports antibiotics from a variety of classes including ~-Iactams (e.g. oxacillin), macrolides (e.g. erythromycin), fluoroquinolones (e.g. ciprofloxacin), oxazolidinones (e.g. linezolid) and tetracyclines (e.g. minocycline). The extent of AcrA/AcrB/TolC mediated activity on ~-Iactam resistance positively correlated with the Iipophilicity of the side chain [29].

""

QP

^\,. ^A ^"'-

~ I ^H,N ^6"

""- ^OCH,

do

""- OCH_J ~,j ~

Acriflavine Rhodamine 6G Berberine Tetraphenyl- Ethidium phosphonium

"~Q

^o^!,,""' ... OH 0

err ^p

F x X f o H

N, /i ~ ~ " N/

\ - j 0

o)=t -jc : '

^OH ^HN

IN

^#' ^N

Or

c6

HN~

L

₀

Ciprofloxacin Erythromycin Oxacillin

HO'\\~'

Taurocholate Dodecylsulfate

Phe-Arg-~-naphthylamide (PA~N)

/ 'g

~

Hexane

NH~

OH 0 0 6" 6"

Minocycline N-Phenylnaph- thylamine (NPN)

Cyclohexane

o

Indole

1-(1-naphthylmethyl)-piperazine (NMP)

Fig. (3). Substrates and inhibitors of the AcrAB-ToIC effiux system. The system confers resistance to a wide variety ofnoxiolls substances like dyes, di fferent classes of antibiotics, detergents, bile salts and small organic molecules. Phe-Arg-p-naphthylam ide and 1-( 1- naphthylmethyl)-piperazine (NMP) inhibit RND/MFP/OMF effiux systems.

(5)

The question of the physiological role of these efflux systems in Gram-negative bacteria is still a matter of debate.

It has been proposed that the efflux of bile salts allows the survival of Gram-negative bacteria in the gut [31], or that such efflux systems relieves the cell of toxic metabolites such as indole or fi'om products of membrane and murein recycling [32, 33J. More recently, the AcrA/AcrB/TolC efflux'system was postulated to transpor~ molecules mediating signals for quorum sensing [34]. An important question arises with respect to the broad substrate specificity and how the efflux system prevents transport of essential hydrophobic cofactors (e.g. flavins) or phospholipids out of the cell. An- other line of investigation is the specific inhibition of these transporters by non-antibiotic compounds. Quite recently, inhibitors of the AcrA/AcrB/TolC and the MexA/MexBI OprM efflux systems like Phe-Arg-p-naphthylamide or aryl- piperazines have been postulated to inhibit the efflux of other substrates by high affinity competitive binding to the drug binding pocket [35, 36].

Transport Across Two Membranes

The contribution of RND/MFP/OMF efflux systems which span the two membranes of the Gram-negative cell to the intrinsic drug resistance is remarkable despite low expression of the pump constituents when compared to expression of drug pumps like the tetracycline transporter Tet, which appears to act as a single component in the inner membrane [37J. The ability of the RND/MFP/OMF efflux system to capture drugs from the inner membrane and the direct transport into the extracellular medium bypassing the periplasm results in a synergistic resistance effect [28J. In contrast, pumping activity of the inner-membrane transporter Tet would cause accumulation of tetracycline in the periplasm from where it easily diffuses over the cytoplasmic membrane back into the cell [3 I]. A recent report suggests that the tetracycline resistance phenotype by Tet is dependent on the presence of the AcrA/AcrB/TolC system [38]. The authors postulate Tet catalyzed transport of tetracycline into the periplasm, from where AcrA/AcrB/TolC pumps the antibiotic over the outer membrane.

Clinical Relevance of Tripartite Efnux Systems

RND/MFP/OMF efflux systems are associated with a major threat to human health as they playa central role in multiple antibiotic resistance of pathogenic Gram-negative bacteria such as Campylobaeter jejuni, E. coli, Klebsiella pneumoniae, Aeinetobaeter baumannii, p, aeruginosa and Salmonella enteriea serovar typhimurium (reviewed in [39- 41)). Highly resistant strains usually arise due to higher expression of the efflux pumps caused by mutations of regula- tory proteins directly acting as repressors of transcription of the MFP/RND loci (e.g. AcrR) or indirectly by repressing the expression of upstream transcriptional activators (e.g.

MarR repressing MarA) [42]. The presence of the efflux pump(s) facilitates the selection for high-level resistance through chromosomal mutations leading to target modifica- tion [40].

Outer Memb"ane Factor (OMF) Tole

The laIC (tolerance to Colicins) locus of E. coli was iden- tifled via a mutation causing a Colicin E I resistant pheno-

type. This mutant is also highly susceptible to bile salts and dyes [43J. TolC is postulated to act as Colicin EI entry pore [44]. It also forms the outer membrane pore for HlyA toxin export by the type I secretion system in concert with an ABC transporter, HlyB, located in the inner membrane, and the membrane fusion protein HlyD [45]. Moreover, TolC acts as the OMF of all described RND/MFP/OMF efflux systems of E. coli (see above). Structures of several OMF members, TolC of E. coli, OprM of P. aeruginosa and VceC of Vibrio eholerae have been solved by X-ray crystallography at 2.1

A,

2.6

A

and 1.8

A

resolution, respectively [46- 48] (Fig. 4). TolC and its homologues form homotrimers presenting a 40

A

OM located, pore forming [3-barrel domain (inner diameters vary between 6 -13

A

in the different structures) and an a-helical domain protruding 100

A

into the peri plasm, which is closed at the proximal end. Every monomer contributes 4 anti parallel p-sheets constituting the p-barrel domain of the trimeric structure and 4 a-helices, which pack in an left-twisted anti parallel arrangement to form the long periplasmic conduit [46]. The closed pore is known to open upon recruitment by substrate-laden complexes in the inner membrane during HlyA toxin export [45].

The transition to the open state of TolC is proposed to be achieved by an iris-like realignment of the helices at the tunnel entrance [49]. Mutants where intra- and intermolecular hydrogen bonds and salt bridges at the proximal end were abolished appeared to be completely functional with respect to HlyA export, but showed increased single channel con- ductance in black lipid bilayers [49]. Introduction of engi- neered disulfide linkages to avoid the opening of the proximal pore resulted in TolC mutants causing decreased drug and HlyA transport capacity [50]. In a screen selecting for TolC mutants causing decreased drug resistance, mutations resulted in a constitutively open TolC channel and were leading to influx of large antibiotics such as rifampin and vancomycin [51]. Further and new insights on TolC structure and function, described by B. Luisi and colleagues, can be found in this issue.

Membrane Fusion Protein (MFP) AcrA

The membrane fusion protein (MFP) family was given its name because of considerable sequence homology to the membrane fusion protein (F protein) of paramyxovirus 5 [52]. Bacterial MFPs are essential partners for RND/MFPI OMF efflux systems but also work in concert with transporters of the ATP binding cassette (ABC) superfamily [19,45, 53] and members of the major facilitator superfamily (MFS) [28]. Two structures of bacterial MFPs, MexA of P. aerugi- nasa and AcrA of E. coli, have been solved by X-ray crystallography [47, 54, 55] (Figs. 4 and 7). In all three crystal structures (MexA was solved independently by two research groups) at resolutions between 2.3

A

^and3

A ,

^on^ly^approxi-

mately two thirds of the amino acid residues of the MFPs could be assigned. A short N-terminal tail and the C- tenninal part of MexA (residues of unprocessed MexA 283-383 [54] or 302-383 [47], respectively) were highly disordered within the crystal or, in case of AcrA, had to be removed in order to obtain crystals. Sensitivity to proteolytic digest further supported the notion of the high flexibility of the C- terminal part [55, 56]. The structures of the MexA and AcrA core proteins are divided into three parts, the p-barrel do-

(6)

30A Tole docking domain

TolC Extracellular

Periplasm

a-helical hairpin

Fig. (4). Stmctures of the proteins constituting the tripartite AcrAB-ToIC efflux machinery. AcrB (PDB entries: IIWG, 2GIF) resides in the inner membrane (1M) and is composed of the transmembrane domain, the porter (pore) domain and the TolC docking domain. TolC (PDB entry: I EK9) is integrated into the outer membrane (OM) with its p-barrel domain and fonns a long conduit in the periplasm by its a-helical domain which narrows to a closed entrance at the proximal end. AcrA (2F I M) is divided into the p-barrel domain, the lipoyl domain and the a-helical hairpin. Its N- and C-tennini are not assigned due to missing electron density. AcrA is associated with the inner membrane via a N- terminally attached lipid anchor Adapted from Eswaran e/ at. 2004.

main, the lipoyl domain and the a-helical hairpin. Its N- term inal signal sequence is responsible for the translocation of AcrA and MexA into the periplasm and is subsequently cleaved producing the mature protein. AcrA and MexA are attached to the inner membrane via a lipid anchor (palmitoy- lation) linked to a cysteine residue at the N-terminus of the mature protein (C25 and C24 of unprocessed AcrA and MexA, respectively). However, the membrane anchor appears not to be essential for drug efflux activity, as expression of soluble, non-membrane attached, periplasm ically located mutants of the AcrA and MexA leads to a normal resistance phenotype [57, 58]. AcrA was suggested to exist as a trimer, based on cross-linking data and blue-native gels, respectively [59,60]. MexA has been crystallized as a higher aggregate (tridecamer aggregate in the asymmetric unit) [47, 54] from purified samples containing mainly monomeric [47, 54] but also dimeric MexA (reported by Akama el af. [47]).

Resistance Nodulation Division (RND) Pump ACI'B RNO proteins are the actual pumps of the tripartite RNO/MFP/OMF efflux system, since both substrate specificity and energy transduction is attributed to this inner-

membrane component. Although for many RND pumps the substrate specificity has been characterized, most of the structural and functional experiments were carried out on MexB of P. aeruginosa and AcrB of E. coli.

Stmctul'e of AuB

The structure of trim eric AcrB has been initially solved at 3.5

A

resolution in space group R32 with one monomer in the asymmetric unit [61-63] (Figs. 4 and S). The AcrB monomer consists of 1049 amino acids and exhibits sequence homology and similar structural architecture between its N- and C-terminal half, indicating an early gene duplica- tion event [52] (Fig. 6). Interestingly, co-expression of the individual N- and C-terminal halves of the AcrB homologue MexB in trans (Residues 1-510 and 511-1049, AcrB numbering) resulted in an active RNO-pump [64). Full length AcrB contains twelve transmembrane a-helices constituting the transmembrane domain (Fig. SA, D). TM4 and TM 10 are surrounded by the other transmembrane helices of the monomer and harbour the residues 0407, 0408 (TM4) and K940 (TM I 0) (Fig. 5D), which appear to play an essential role in proton translocation [65, 66]. The connecting loops

(7)

A

Tole docking domain

Pore

Trans-

memb,·u, .. . --..

domain

c ^D

Fig. (5). A closer look to AcrB. (A) Side view. Trimeric AcrB is composed 01' the TolC docking domain exhibiting a Funnel-like structure narrowfng to a central pore located in the porter (pore) domain and the transmembrane domain confining a central cavity. (8) Top view. The trimeric state 01' AcrB is mainly stabilized by the intennonomer connecting loops. A central pore is Fonned by three a-helices, donated by the PN I subdomains of each AcrB monomer. (C) Top view onto the porter domain with its subdomains PN I, PN2, PC I and PC2. The lateral cletl between subdomains PC I and PC2 is suggested to accommodate AcrA, Murakami e{ at. 2002. (D) Top view onto the transmembrane domain. TM4 and TM 10 are sUITounded by other 10 transmembrane helices. The transmembrane helices 0 I' the monomers con Fine a large central cavity.

between TM I and TM2 as well as TM7 and TM8 fold into two large periplasmic domains (Figs. SA and 6) which account for approximately half of the amino acids of the full length AcrB molecule. The AcrB monomers form a trimer which appears to be stabilized by the intermonomer connecting loops protruding from the TolC docking domain into the adjacent monomer (Fig. 58). The periplasmic part of AcrB consists of the TolC docking domain (divided into the DN and DC subdomains), which is located most distant from the

membrane plane (Fig. SA), and the porter (pore) domain.

The TolC docking domain exhibits a funnel-like structure narrowing to a central pore located in the porter domain, the latter composed of subdomains PN I, PN2, PCI and PC2 (Figs. SA,C and 6). The characteristic central pore is formed by three a-helices (designated pore helices), donated by the PN I subdomains of each AcrB monomer (Fig. S8,C). The pore has a small diameter and therefore does not allow drug passage in this conformation. The PCI and PC2 subdomains

(8)

TolC

Docking Domain

Pore Domain

Peri plasm

Transmembrane Domain

Cytoplasm H,N

n n

Fig. (6). AcrB monomer secondary structure scheme (residues 2-1033), The porter (pore) domain consists of four sl!bdomains, PN I, PN2, PC I and PC2, The TolC-docking domain has two subdomains, DN and DC. TM, transmembrane helices; Na, NP, Ca and Cp are a-helices and p-sheets oFthe N-terminal part or the C-terminal part of the periplasmic domain, la2 is the cross-a-helix at the cytoplasmic side, N- and C-tenninal halves are depicted in blue and magenta, respectively, The intermonomer connecting loop from the adjacent monomer is depicted in green, Adapted From Murakami el aI., 2002 and extended with inFormation drawn from the asymmetric AcrB structure (PDB ently: 2GIF), constitute a cleft at the periplasmic periphery or the porter

domain which is suggested to accommodate AcrA (Fig. SC) [61]. At the interior of the protein near the membrane plane, the central pore leads to a central cavity and further to a 30- 35

A

wide, presumably lipid-tilled transmembrane hole de- tined by the ring-like arrangement of the TM helices of the trimer (Fig. SA,D). Three vestibules at the monomer inter- face located just above the membrane plane lead towards the central cavity (Fig. SA,D).

Recently, we [67] and others [68, 69] solved a structure of AcrS which does not entail three-fold symmetry of the trimer, In contrast to the symmetric structure, this asymmetric structure reveals three different monomer con formations representing consecutive states in a transport cycle, loose (L), tight (T) and open (0) (see below). The structural changes in the T monomer create a hydrophobic pocket, which is not present in the other monomers. We [67] assumed that this pocket is a substrate binding pocket inside the porter domain and Murakam i and colleagues [68] showed binding or minocyclin, 9-bromo-minocyclin and doxorubicin to this pocket (see below).

Substrate Recognition

The first symmetric structure of AcrS solved at 3.5

A

was reported to be ligand free [61]. Another study from a different research group reported on four structures of symmetric AcrS-substrate complexes derived from diffraction data of crystals at 3.5-3.8

A

resolution [70). Enforced by the three-fold symmetry of the crystal space group R32, three densities, which were interpreted as drug molecules binding to the AcrS trimer, were located in proximity to F386, F388, F458 and F459 in the central cavity. It was suggested that these structures represent the first binding step in the catalytic cycle before the drugs are further transported through the central pore towards TolC [70]. The interpretation of these results are, however, not beyond doubt due to the low resolution, the high R (and Rrlee)-factors and the sub-stoi- chiometric drug to AcrB molar ratio of 0.2 used in the co- crystallization assays. We described co-crystallization and crystal soaking experiments under very similar conditions of the same and other drugs at 10 to 100 times higher drug to AcrS ratios, but did not yield any indication of drug binding despite 2.8-2.9

A

resolution electron density map surveys

(9)

[63]. Yet other crystal structures (obtained from crystals grown in R32 space group) by Yu el al. [71] showing ligand binding to a functionally active N 109A mutant of AcrB.

Drug binding was interpreted to occur at the lateral cleft near the hydrophobic residues F664 and F666. Data from the same crystals also included densities near the central cavity as has been reported earlier by the same group [70]. The derived structures were presented including 6 drug molecules bound to the AcrB trimer. Alanine substitution of F664 and F666 within the proposed periplasm ic binding pocket led to substantial decrease in MIC values wheras on the other hand, functional characterization of mutants with alanine substituted F386, F388, F458 and F459 within the central cavity only showed marginal effects on efflux activity [71]. The latest published AcrB structure (derived from R32 crystals diffracting to 3.5 A) by TOl'llroth-Horsefield el al. [72] includes a co-crystallized a-helical peptide, which was identified as a fragment of Yajc. This structure was published with six ampicillin molecules (originating from the selective growth medium) located inside the central cavity of the trimer despite the fact that the protein was purified and crystallized in the absence of this substrate.

Analysis of our R32 data at 2.7

A

[63] showed that these crystals suffered from merohedral twinning [67]. A high twinning fraction results in blurred electron densities and hence makes interpretation of potentially bound substrates very difficult. Moreover, AcrB has been crystallized in the presence of dodecylmaltoside and cyclohexyl-hexylmalto- side, both known substrates of AcrB. The concentration of detergents inside the protein crystals is estimated to be approximately 20% (DDM: 392 mM) and would present seri- ous competition for other substrates present at low or sub mM concentrations. It should therefore be considered (espe- cially with low resolution data) whether observed densities are derived from twinning artefacts and if not, whether the observed densities could represent sugar headgroups of the detergent attached to the protein. One exemplary approach to assign observed densities to bound substrates has been given by Murakami el af. [68]. This research group deployed 9- bromo-minocyclin to make use of the anomalous signal pro- duced by Br at 0.919

A

and in this way elegantly showed the binding of 9-bromo-minocyclin to AcrB at relative low resolution (3.6 A). Binding of minocyclin, 9-bromo-minocyclin and doxorubicin to the asymmetric AcrB trimer occurred at a completely different site compared to the binding sites of the symmetric AcrB trimer at only one of the three monomers (T monomer) inside the porter domain. Strikingly, no substrate binding was observed in the central cavity of the asymmetric AcrB trimer.

Domain swapping experiments between RND-pumps with distinct substrate specificities (AcrB/AcrD, MexB/MexY, AcrB/MexB) already suggested that the substrate specificity is determined by the large periplasmic loops [73-75]. When both periplasmic loops of AcrD were exchanged with the corresponding amino acids of AcrB, the chimeric protein transported the cognate substrates of AcrB but not those of AcrD. On the other hand, exchange of the TM2 to TM6 and TM8 to TM 12 of AcrD with the corresponding residues of AcrB did not result in an AcrB transport phenotype [73]. The same result was obtained for the respective chimera of MexB and MexY. In addition, one by one exchange of TM I to

TM 12 of MexB by the respective TMs of Mex Y always lead to chimeric transporters exhibiting substrate specificity of MexB [74]. Sequential replacement of the AcrB C-terminus by corresponding residues of MexB or the MexB N-terminus by the respective amino acids of AcrB suggested, that the N- terminal periplasmic loop between amino acids 60 to 612 is responsible for substrate specificity [75]. Six random mutants of MexD conferring increased resistance to carbenicil- lin, a poor substrate of wild type MexD, were mapped to both periplasmic loops [76]. These results are all in agree- ment with the key role of the periplasmic loops on substrate specificity.

Elevator Mechanism

Based on the symmetric AcrB structures derived from R32 crystals, it was hypothesized that diffusion of substrates occurs from the cytoplasmic site of the membrane via the transmembrane domain along a groove between TM8 and TM9 (for completely hydrophobic substrates except p- lactams) or from the periplasmic site of membrane via the vestibules (for p-Iactams) into the central cavity. Subse- quently, opening of the central pore would allow the transport of the substrates through the pore, to TolC and export to the extel'llal medium. Large conformational changes were postulated to be associated with this transport function, which was tentatively termed "elevator mechanism" [13]. Systematic cysteine scanning of the pore helix highlighted the importance of the pore helix side chains protruding into the central pore. The activity of AcrB pore mutants was sub- stantially decreased due to intermolecular disulfide bridge formation between AcrB monomers [77]. However, a satis- factory interpretation of the latter results is also obtained in the context of the new structural [67,68] and functional data [78], as will be discussed below.

Energy Transduction

Members of the RND superfamily are proposed to function as proton/substrate anti porters [2]. One of the most con- vincing experiment supporting this idea was carried out with reconstituted heavy metal-ion transporter CzcA, a member of the heavy metal efflux (HME) family (TC#2.A.6.1). Besides an observed fast facilitated diffusion of Zn²+ ions, a much slower proton/Zn2+ anti port could be measured, which disap- peared upon the mutation of D407 or E414 (AcrB numbering) [12]. These residues are presumed to playa central role in proton translocation in members of the HAE I family (such as AcrB and MexB) as well [65, 66]. Reconstituted AcrB in proteoliposomes has been shown to transport fluo- rescent phospholipids from donor to acceptor vesicles in the presence of a pH gradient. Addition of unlipidated, soluble AcrA facilitated this transport. Dissipation of the artificially generated proton gradient across the AcrB containing liposome membrane was accelerated in presence of substrates, indicating AcrB-mediated H+/drug antiport [79]. Re- constituted AcrD, a close homologue of AcrB, which transports rather hydrophilic aminoglycosides such as gentamycin, was able to trans locate gentamycin in exchange for protons in presence of AcrA [80]. Systematic mutational analysis of all transmembrane domain localized, titratable residues of AcrB and MexB revealed that only five residues, D407,

D408, K940, R971 and T978 (AcrB numbering) were essen-

(10)

tial for drug transport [65, 66]. These residues were suggested to constitute a charge network playing a central role in proton transduction, a notion supported by the AcrB structure [61,67-69].

The Tripartite RND/MFP/OMF Efflux System Forms a Multiprotein Complex

The elimination of one gene of constitutively expressed tripartite RND/MFP/OMF efflux systems leads to a drug sensitive phenotype in Gram-negative bacteria. A growing number of publications deal with the mutual interaction of the single components within the complex.

Biochemical Evidence

Physical interaction between AcrA and AcrB [56, 59, 81], AcrA and TolC [51, 56, 81], AcrB and TolC [56, 81]

and MexA and MexB [82] could be demonstrated using the cleavable cross-linker dithiobis(succinimidylpropionate) (DSP). Cross-linking efficiency between AcrA and AcrB was not dependent on the presence of TolC [56, 59, 81], nor were cross-links between AcrA and TolC depending on the presence of AcrB [51, 56]. Moreover, interaction between AcrB and TolC was also detected in absence of AcrA [56, 83]. Using hetero-bifunctional cross-linkers, TolC containing single cysteine mutations on the lower a-helical domain close to the entrance aperture could be cross-linked with ActA Reciprocally, site-specific cross-linking of AcrA cysteine variants to wild-type TolC identified the N-terminal a- helix of the a-helical hairpin of ActA to be involved in the interaction with TolC [84]. Direct interaction between AcrB and TolC was demonstrated using disulfide cross-linking of cysteines located on distal loops of the TolC docking domain of AcrB with cysteines introduced at the proximal part of TolC [83]. The mutual interaction of the components of the tripartite complex does not depend on the presence of added substrates [51, 81, 83], or the proton-motive force (pm f) [56, 81]. Moreover, the formation of the tripartite complex is not affected by mutations within proton translocation site 0 f AcrB, which render the protein pump inactive [81, 83]. The tripartite MexA/MexB/OprM complex was co-purified from overproducing E. coli cells without the use of cross-linking agents by cobalt-chelate affinity chromatography. The overproduction of all three components of the tripartite efflux machinery was crucial to achieve co-purification of the complex [85]. Physical interaction between ActA and TolC and between AcrA and AcrB was shown in vitro by isothermal titration calorimetry, whereas AcrB and TolC do not show interaction in this particular assay. Because of rather complex titration patterns non-linear regression of the data yielded two and four different dissociation constants for the interaction between AcrA and AcrB and between ActA and ToIC, respectively. The multiple Kd values are proposed to represent sequential phases of the respective protein-protein interactions and range between 0.35 pM and 17.2 ~IM [56].

Genetic Studies

Genetic studies were in further support of the biophysical data described above and in addition were successfully exploited to map potential interacting domains between the components of the efflux system. MdtE (formerly YhiU) is

an MFP of E. coli which operates in conjugation with TolC and MdtF, but not with AcrB (see Table I). Chimeric analysis between both MFPs AcrA and MdtE revealed that a re- gion within the C-terminal part of AcrA (residues 290-357, numbering of unprocessed AcrA throughout) interacts with AcrB [22]. In the AcrA crystal structure, most of this part (residues 300-357) is not resolved due to disorder [55]. A screen for rescue mutations of an inactive TolC mutant (P246R, S350C) within the acrRAB regulon yielded colonies with mutations in AcrA (10 cases) and AcrR (one case), but not in AcrB [86]. The TolC P246R_S350C double mutant was shown to be partially defective in assembly, but in the presence of some of the selected AcrA mutants, its proper integration into the outer membrane was improved, i.e. it was stabilized in presence of mutant AcrA presumably due to a direct physical interaction. Interestingly, this assembly improvement was fully dependent on AcrB, because it could not be reproduced in an acrB negative background. All but one ActA mutation map on the ~-barrel domain [86] (Fig. 7).

Mutations in MexA compromising antibiotic efflux of the MexA/MexB/OprM system mapped in the ~-barrel domain of MexA as well, but were demonstrated to lead to an interaction loss with MexB. However, the effect of the mutations on OprM interaction/assembly was not tested [82] (Fig. 7).

Screening for MexA mutants interplaying with OMF member OprN, which is not a cognate partner of MexA, revealed a mutation in the a-helical hairpin of MexA suggesting that the latter domain is involved in interactions with the OMF (Fig. 7). This suggestion is even more appreciated since substitution of the a-helical hairpin of MexA with the corresponding residues of MexE (which is the cognate partner of OprN) results in a functional heterologous MexNMexBI OprN efflux system [87]. Similarly, the substitution of the AcrA hairpin subdomain with that of MexA is sufficient to change the heterologous AcrA/AcrB/OprM hybrid efflux system from a non-functional into a functional export ma- chine [88]. In a directed evolution approach, mutant TolC variants carrying single or double amino acid substitution were selected to form an active heterologous MexA/MexBI TolC efflux pump. All substitutions leading to an active pump map to the periplasmic equatorial domain and entrance coiled coils of TolC and converge to the amino acid sequence of the native MexA/MexB partner OprM [89] (Fig.

7). Gain-of-function mutations within VceC resulted in functional heterologous AcrA/AcrBlYceC export machinery and mutations map as well at the entrance coiled coils of VceC. However, in this case no convergence towards the amino acid sequence of TolC could be observed [90]. In summary, the current data support the motion for a physical interaction of the Ml~P's a-helical hairpin with OMF's coiled coils below the equatorial domain at the periplasmic tip.

[n contrast, interactions between RNDs and MFPs are still poorly understood. MexB mutations resulting in an antibiotic-sensitive phenotype mapped at the lateral cleft of the periplasmic domain of MexB. Several mutations on the ~

barrel domain of MexA suppress the MexB mutants, restore their pumping activity and confer resistance towards their host [91]. Very recent work [92] on the compatibility between MexB of P. aeruginosa and AcrA/TolC of E. coli sur- prisingly showed functional complementation for dodecylsulfate, novobiocin and ethidium, but not for other MexB

(11)

A B c

Fig. (7). Visualization of mutations in AcrA (A) and MexA (8) related to functional interactions with the RND and the OMF components of the tripartite efflux system. The p-barrel domain, I ipoyl domain and the a-helical hairpin are colored in blue, yellow and red, respectively.

(A) Substitutions in AcrA that suppress a drug sensitive phenotype caused by mutant TolC (P246R_S350C) are shown as space fill representations and coloured in aquamarine (Gerken e/ al. 2004). (8) MexA residues which substitution caused a drug sensitive phenotype of MexAB-OprM, are shown as space fill representations and coloured in aquamarine (Nehme e/ al. 2004). Substitutions of residues involved in the rescue ofa defective MexB mutant are coloured in light orange (Nehme e/ al. 2005). The substitutions ofQ 116R in MexA allowed functional interaction between MexA and the non-physiological OMF partner OprN and is indicated in a pink space filled representation (Eda e/

al. 2006). (C) Substitutions of residues in TolC that afforded an active MexAB-ToIC efflux system are depicted as aquamarine space fill representations (Bokma e/ al. 2006).

substrates. Single substitutions in AcrA or a double mutation in MexB (T329I1A802V) were sufficient for improvement of the functional alignment between the two non-cognate subunits demonstrated by the markedly elevated MIC values of these mutants.

I" sitico Modelling

The availability of structures of all three individual components of the tripartite RND/MFP/OMF efflux system was exploited to predict the tripartite structure of the AcrAIAcrBI TolC transport machinery. All models propose an assembly of trim eric AcrB and trim eric TolC co-axially along their three fold symmetrical axes (as in Fig. 2) but differ in the AcrA:AcrB:ToIC stoichiometry. Because MexA was crystallized as a tridecameric ring-like structure, a ring of 9 MexA monomers enclosing a MexB-OprM complex was suggested, i.e. a 3:1:1 stoichiometry [54]. Another model favours a 2: I: I stoichiometry between MexA, MexB and OprM, with three MexA dimers oriented in a ring-like arrangement towards Mex B-OprM [47]. Based on calcu lations and from disulfide cross-link studies, an analogous model was suggested for the AcrAIAcrB/TolC complex [88]. In support, the 2: I: I stoichiometry would account for the abundance 0 f MexA relative to MexB and OprM in the P. aeruginasa cell, as quantified by immunoblotting [47]. In yet another in silica model, a I: I: 1 stoichiometry was suggested for the AcrAl AcrB/TolC complex, which in contrast to the studies men- tioned above permits direct contacts between AcrB and TolC

[48]. The 1:1 stoichiometry between TolC and AcrA is sup-

ported by site-specific cross-linking data of single cysteine substituted TolC to AcrA and vice versa [84]. The experimental data allowed a data-driven docking approach to

model the interaction surface between AcrA and TolC [84].

All models described above predict the interaction between the a-helical hairpin of the MFP and the coiled coils orthe a- helical domain of the OMF and in addition, one study [48]

predicts an interaction between the ~ -barrel of the MFP with the lateral cleft of the RND pump. A central role within the tripartite complex inheres to the MFP. Molecular dynamic simulation using the structure of MexA as a starting point revealed correlated movements between the ~-barrel domain and the a-helical hairpin, which are suggested to provide considerable flexibility within MexA, enabling it to dynami- cally bridge the RND and OMF [93]. The calculated movements would result in a hinge bending of about 19° at the junction between the a-helical hairpin and lipoyl domain (Figs. 4 and 7). Interestingly, comparison of the four AcrA monomer structures (PDB 10: 2F 1 M) in the asymmetric unit of the AcrA crystal structure reveals a maximal difference of that hinge angle of about 15°. Despite all efforts, the exact stoichiometry and assembly of the tripartite RND/MFP/OMF efflux system remains elusive and crystal structures of the respective complexes have to be solved to get clearer evidence.

DISCUSSION

The Protein Data Bank (www.pdb.org) contains at the moment of writing 24 AcrB structures. The first AcrB structure (year 2002, PDB entry: IIWG) [61] was based on 3.5

A

X-ray diffraction data from R32 crystals containing one un- liganded AcrB monomer in the asymmetric unit i.e. it describes a symmetric AcrB trimer. In the following years, other groups deposited symmetric (wildtype and mutant) AcrB X-ray structures (based on 3.1-3.8

A

data) with and

(12)

Fig. (8). Schematic representation of the AcrB alternating site functional rotation transport mechanism. The conformational states loose (L), tight (T), and open (0) are colored blue, yellow and red, respectively. (A) Side-view schematic representation of two of the three monomers of the AcrB trimer. AcrA and TolC are indicated in light green and light purple colors, respectively. (8) The lateral grooves in the Land T monomer indicate the substrate binding sites. The different geometric forms reflect low (triangle), high (rectangle), or no (circle) binding affinity for the transported substrates. In the first state of the cycle, a monomer binds a substrate (acridine) in its transmembrane domain (L conformation), subsequently transports the substrate from the transmembrane domain to the hydrophobic binding pocket (conversion to T conformation) and finally releases the substrate in the funnel toward TolC (0 con fornl ation). AcrA is postulated to participate in the transduction of the conformational changes from AcrB to ToIC, which results in the opening of the TolC channel and the facilitation of drug extrusion to the outside of the cell. Adapted and modified fi'om Seeger el al., 2006.

without ligand [70-72, 94, 95]. With one exception [72], most of these structures were derived from the first structure (POB entry: IIWG) taken as a template. In the years 2006 and 2007, 6 structures describing an asymmetric AcrB trimer were deposited (POB entries: 2D1-IH, 20R6, 20RO, 2GIF, 2I-1RT, 2J8S)[67-69]. The best resolution structure (2.5

A,

POB entry 2J8S) was obtained with AcrB/designed ankyrin repeat proteins (OARPins) co-crystals [69]. Two of the deposited asymmetric structures were derived from 3.1 and 3.3

A

^d^ata of AcrB/minocyclin and AcrB/doxorubicin co- crystals, respectively [68]. From the asymmetric structures, a model for drug transport based on conformational cycling of the monomers by the RNO pump AcrB has been proposed (Fig. 8) [67, 68].

Conformational Cycling and Binding Change Mecha- nism

Visual inspection of the periplasmic domain of the asymmetric AcrB structure (POB entry: 2GIF) -viewed from the periplasmic side perpendicular to the membrane plane- and direct comparison to the structure of the

alP

subunits and y subunits of bovine FIFo ATP synthase (PDB entry: I BMF [96]) -viewed from the cytoplasl1l- leads to the overwhelm-

ing impression that the structures are analogous (Fig. 9). The y rod of the ATPase is ti Ited towards one of the three alp subunits. Likewise, in one monomer of the AcrB trimer the pore a-helix (Na2, Fig. 6) is tilted towards the neighbouring monomer's PN2 subdol1lain. In the ATPase and AcrB structures, the monomers are in distinct different conformations, des'ignated loose (L), tight (T) and open (0) I. Long before structural details of the FIFo A TP synthase were known, Paul Boyer postulated the binding change mechanism for this enzyme, describing a conformational cycling of the (a and)

p

subunits through the states loose, tight and open, leading to the synthesis of ATP [97-99].

One of the three key points of his binding change mechanism is that the energy input (i.e. proton motive force) is not used to form ATP, but to release the ATP molecule from the

I There is a marked difference between the designation of the uJp subunits of bovine F, F. ATP synlhase and Ihal or Ihe AcrB monomers. The 1 rod or Ihe ATPase is lilted towards the "open" alP subunits (despite appearances in Fig. 9), whereas the P}.J1 subdomain of the "open" monomer of AerB is tilted towards the "tight" monomer. The functional relevance of the y rod of the ATPase as energy transducer is more likely comparable wilh TM8 or AcrB. The coil 10 helix lransilion ofTM8 in Ihc AcrB "open"

monOlller appears to cmphasize its rolc as energy transducer.

(13)

Fl ATPase, bovine mitochondria

Loose, L

Tight, T Open, 0 Tight, T

AcrB, E cherichia coli

Loose, L

Fig. (9). Structural analogy between the

alP

subunits and y subunits of bovine FIF_oATP synthase (POB entry: I BMF, Abrahams el a/., 1994) (left, viewed ITom the cytoplasm) and the periplasmic domain of the asymmetric AcrB structure (POB entry: 20110, Seeger el al., 2006) (right, viewed from the periplasmic side perpendicular to the membrane plane). The structures are presented as ribbon diagrams and the designation of the individual monomers is indicated (Loose (L), Tight (T) and Open (0». See text for details.

~ subunit. The second key'point is the catalytic cooperativ- ity: ATP can only be released when ADP and Pi are bound to another monomer within the hexameric F I part of the A TP synthase (so called bi-site activation). The third key point is the rotational catalysis.

In analogy to the binding change mechanism (also called the alternating site mechanism) our group [67] and Mura- kami et al. [68] independently postulated an analogue alternating site functional rotation mechanism for the transport of drugs by AcrB (Fig. 8). The general concept describes a functional rotation starting with the (loose) binding of substrate to a low affinity site on the L monomer, followed by conformational change to the 'I' conformer and tight binding of the substrate in the designated binding pocket and finally conversion to the 0 monomer resulting in the release of the substrate towards Tole. The 0 monomer finally converts to the L conformation restarting the cyclic event (Fig. 8). It has to be pointed out that the direction of events e.g. conversion from L toT, T to 0, and back to L is based on interpretation of structural data only and that further experimental proof is needed to establish the consecutive conformational states. To date, three considerably different conformational states of the AcrB trimer have been crystallized: The "all-loose" or LLL conformation derived from dirti'action data of crystals grown in the R32 space gr.oup [61], the asymmetric or L TO conformation obtained from crystals lacking the three-fold symmetry [67, 68] and the "all-tight" or TTT conformation (Eicher, Pos, unpublished). Moreover, cross-linking data supports the conformational tlexibilty within the AcrB trimer to form the LLT, L 1"T and 1"1"0 conformations [78]. As is indicated in Fig. (10), these conformational states have been incorporated into a more detailed scheme for drug transport, taking into account the cooperalivity or bi-site activation, in analogy to the binding change mechanism by the F IF ⁰A TP synthase. In the proposed scheme (Fig. 10), the LLL conformation of the AcrB trimer describes a state when the pro-

tein is depleted from substrate ("resting state" [94]). Binding of substrate to the L monomer results in the conversion from L to 1" and, according to the binding change mechanism, a second substrate has to bind to another monomer as a pre- requisite for the release of the substrate to the outside i.e.

substrate has to bind to another monomer before conversion of the substrate-occupied 1" monomer to the 0 monomer can occur (Fig. 10). The conformational change from the T to the

o

monomer is proposed to be an energy dependent step, in analogy with the energy dependence of the release of A TP from the ~-subunil. Proton uptake from the periplasm is an- ticipated in this step (Fig. 10). Structurally, the conversion of the 1" monomer to the 0 monomer might be sterically facilitated by the formation of the adjacent T monomer (Figs. 10 and II). The conversion from the L monomer to the 1"

monomer comprises substantial movement of the PN2 subdomain (L TO to 1"1"0). This causes the loss of restraints for the adjacent PN I subdomain of the already existent T monomer from the LTO trimer and facilitates its pronounced incl ination towards the other 1" monomer (Fig. II) [67, 68].

Its conversion to the 0 monomer follows the cOl;version from TTO to L 1"1". The combined effect of binding of a second substrate to the AcrB trimer (leading to the formation of a second T monomer) with the binding of (a) proton(s) to the transmembrane domain (driven by the proton motive force), results in the formation of the 0 monomer.

The basis for the conversion of the 0 monomer to the L monomer (TTO to L 1"1") is still unclear (Fig. 10). However, two neighbouring 0 monomers would cause an intermono- meric steric clash between the PN I and PN2 subdomains like has been shown for the 0 and L monomer (Fig. II).

Binding of a third substrate to the L conformer of an L TT AcrB trimer (Fig. 10) is considered to be possible (e.g. in the case of high substrate concentration) and will trigger the creation of the third 1" monomer resulting in the "all-tight"

conformation (TTT) , a conformation found in AcrB crystals

(14)

LTO TTO LTT LTO

t

LLT

t

LLL

Fig. (10). Schematic representation orthe AcrB alternating site Functional rotation transport mechanism extended by postulated intermediate steps. The conformational states loose (L), tight (T), and open (0) are colored blue, yellow and red, respectively. The lateral grooves in the L and T monomer indicate the substrate binding sites. The different geometric fornls reflect low (triangle), high (rectangle), or no (circle) binding affinity for the transported substrates. Both states L TO at the Far lefl and far right are identical to the states shown in Fig. (8). State TTT is postulated to occur at high substrate concentration. The states LLL and LL T are postulated to occur in the absence or at low substrate concentrations. See text for details.

(Eicher, Pos el aI., unpublished). Energy input from the proton motive force leads to proton binding to the transmembrane domain of one of the T monomers, formation of the 0 monomer and finally conversion to the L monomer. Interest- ingly, cross-link data supports the formation of more than one of the same conformer (TTL and LL T) in E. coli membranes [78], whereas in crystallization experiments the formation of symmetric forms of the AcrB trimer (LLL, TTT) appears to be predominant to the formation of asymmetric AcrB trimers in crystals. For the crystalline state, this may be explained by the reduced free energy of symmetric crystal contacts as compared to asymmetric ones.

TU!1nels Guiding the Substrate

Upon examination 1'01' cavities and tunnels in the asymmetric AcrB structure [67, 69] (PDB entries: 2GIF, 2JS8) a tunnel system leading to and away from the hydrophobic

substrate binding pocket became apparent (Fig. 12). The L monomer (Fig. 12A, blue) contains a tunnel (designated tunnel 2) starting at the lateral cleft (subdomain PC I/PC2 inter- face) about 15

A

above the putative membrane plane and protruding into the porter domain. In the T monomer (Fig.

12A, yellow), the formation of an additional tunnel (tunnel I) in the porter domain is apparent with its entrance located at height of the TM8 and TM9 groove, which has been postulated to accommodate substrates originating from the membrane [61, 67-69]. Indeed, good resolution (2.5 A) data [69] shows the presence of a dodecylmaltoside molecule (which is both detergent and AcrB substrate) located in the TM8/TM9 groove of the T monomer. Tunnel I merges with tunnel 2 close to the hydrophobic substrate binding pocket (Fig. 12, inset with bound minocyclin). In the 0 monomer (Fig 12B, red), the lateral opening and tunnels I and 2 are absent due to' the closure of the lateral cleft caused by the

The AcrB Efflux Pump : Conformational Cycling and Peristalsis Lead to Multidrug Resistance